It Matters Who Teaches You (Gruber)

Bastian Gruber is the author of Rust Web Development from Manning and is writing its second edition. The post opens with how much he liked LLMs. School never suited him, and at 37 a tool that produces endless explanations of any topic felt like what he had been missing: he imagines taking notes in class and going home to question a model for hours about the things he was too shy to ask. Even when the answers are wrong, bouncing ideas off one has given him insights he would not otherwise have had, which matters to a remote worker. He had come to see LLMs as a net positive, provided the regulation of data centres, training and use gets sorted out.

Lineage

What changed his mind was a book about meditation, Susan Piver's Inexplicable Magic. Piver writes that who you learn from matters because "you have entered the practice via a certain doorway through which many others have passed", and later asks whether there is a lineage behind your instructor and their instructions: "If so, good, because we don't want any made up bullshit." Gruber applies this directly. An LLM has no lineage and nobody stands behind its answer, so what it offers is made-up bullshit that sounds convincing, and you should learn from people who are free of it.

The people who taught him

He came to software through an electrician's apprenticeship, where having a foreman was not a question. Ideally you were paired with one who fitted your skills and ambitions, and that person "looked you in the eye, had a gut feeling and experience, and decided what you were ready for." He credits teachers, senior engineers and managers with as much of his growth as his own work, including what kind of people they were, and says he wrote the first edition of his book thinking about them.

What goes missing

His evidence for the loss is how sessions feel. Two or three hours with an LLM usually leave him empty and burned out. Pair programming, or a room of people working a problem, leaves him inspired and still working after the day is over. An LLM has no sense of who you are; it interprets text and returns paragraphs that sound right, with no teaching and no connection. He will not tell his children about long sessions with a model, but he will tell them about nights spent with other people. That is why he keeps writing books: he hopes one young engineer will someday say they learned something from Bastian Gruber, and not from statistically generated output.

Reading it

Two claims are tangled together here and they hold up differently. The accuracy claim, that LLM answers are made-up bullshit, is the weaker one, and his own opening concedes the answers usually come close to true. The provenance claim is stronger: an answer with nobody behind it is the absence a-voice-from-nowhere describes, text that imitates thought with no path underneath.

The strongest claim is about teaching as a relationship. The foreman deciding what you are ready for is a teacher holding a model of the student, which no amount of good explanation supplies. It is the learner's side of contributor-poker, where a maintainer's review is an investment in a person rather than in a patch, and it is the apprenticeship that programming-still-sucks and rakyll-coding-agents see disappearing as mundane work goes to agents. The evidence collected at pair-programming points the same way, with the largest benefits for juniors on complex tasks. dont-outsource-learning offers a partial answer from inside the tools, that asking a model conceptual questions beats copying its code, but Gruber's point is that a better question to a model is still not being taught. The drained feeling after long model sessions is reported in agentic-coding-fatigue and know-thine-enemy too.